In recent decades, a decrease in snowfall attributed to the effects of global warming (among other causes) has become evident. However, it is reasonable to investigate meteorological causes for such decrease, by analyzing changes in synoptic scale patterns. On the Iberian Peninsula, the Castilla y León region in the northwest consists of a central plateau surrounded by mountain ranges. This creates snowfalls that are considered both an important water resource and a transportation risk. In this work, we develop a classification of synoptic situations that produced important snowfalls at observation stations in the major cities of Castilla y León from 1960 to 2011. We used principal component analysis (PCA) and cluster techniques to define four synoptic patterns conducive to snowfall in the region. Once we confirmed homogeneity of the series and serial correlation of the snowfallday records at the stations from 1960 to 2011, we carried out a Mann-Kendall test. The results show a negative trend at most stations, so there are a decreased number of snowfall days. Finally, variations in these meteorological variables were related to changes in the frequencies of snow events belonging to each synoptic pattern favorable for snowfall production at the observatory locations.
Snow and ice, together with windstorms, are among the most important meteorological hazards in winter. In fact, heavy snowfall can significantly inconvenience users of the transportation infrastructure and can increase the number of injuries and fatalities related to traffic accidents [
Snowfall detection and measurement represent very difficult problems in modern hydrometeorology. For snowfall detection, it is necessary not only to consider precipitation systems, but also to determine other variables such as the level of snowflake fusion (melting level), in order to discriminate precipitation reaching the ground as rain or snow. Determining the melting level is also critical for hydrologic models, since it determines the basin surface exposed to rain and the possibility of flooding. Ground measurements are complicated owing to detection technology limitations, snow drift and accumulation issues, and error definition [
Usually, the melting level is slightly below the freezing level, which is the height of the 0°C isotherm. Lundquist et al. [
In recent decades, several studies have shown the influences of climate change on water resources availability, precipitation, and temperature [
Synoptic scale patterns responsible for severe snowfall episodes have been used in many mountainous areas to define atmospheric conditions under which these phenomena occur [
In this paper, we analyze synoptic scale conditions favorable for the most important snowfall in the principal cities of the northwestern Iberian Peninsula. Once we establish the synoptic scale patterns, we study the frequency of each and their temporal evolution, as well as the trend in number of days with recorded snowfall. Finally, since the number of snowfall days has decreased, we assess whether this negative trend is attributable to changes in large-scale circulation patterns, to variations of temperature, or to a combination of the two.
The northern plateau region is in the northwest Iberian Peninsula (Figure
(a) Iberian Peninsula. (b) Study area of Castilla y León and locations of its principal cities.
Box and whisker plot for number of snow days/year for each station (Ávila, Burgos, León, Palencia, Ponferrada, Salamanca, Segovia, Soria, Valladolid, and Zamora). The lower end of the box indicates the 25th percentile, the central line indicates the median, and the line above the box indicates the 75th percentile. The error lines above and below the box indicate the 95th and 5th percentiles, respectively.
To classify situations conducive to snow in Castilla and León, we selected the most significant snow events in 10 major cities of the region between 1960 and 2011. The choice of these observation sites was made because most of the regional population is concentrated near these cities, which are also in areas more vulnerable to snowfall. Moreover, long-term, reliable time series are available from stations at these sites. We also selected stations overseen by a professional staff and, where possible, airport weather stations, because these are usually at very stable locations. The criteria for snowfall events selection were days with more than 5 mm of accumulated precipitation and temperature below 2°C. According to Lundquist et al. [
Synoptic conditions of snow events were characterized by extracting temperature and geopotential height at 850 and 500 hPa from NCEP-NCAR reanalysis data [
Synoptic scale patterns were produced from reanalysis data using principal component analysis (PCA) and cluster techniques (CT) based on retained principal component loadings. Days possessing similar loadings on the extracted components were clustered together. Both methods have been widely used for classifying and establishing weather patterns [
PCA is designed for the reduction of the number of variables while maintaining strong representation of variability contained in the original data. PCA is a method that ensures that only the fundamental variation modes of the data are considered in the clustering process [
The CT allows for separation of data into groups, whose identities are previously unknown. In this study, the selected CT was the nonhierarchical
Once the synoptic patterns favorable for the most significant snow events have been identified, we test how frequency variations of snow events in the observed patterns affect the number of snowfall days recorded at each station. To this end, we analyzed trends in the number of snow days (days on which solid precipitation occurred during the study period) between 1960 and 2011 at the stations used previously to extract the synoptic patterns. We further investigated trends in precipitation and temperature at the selected stations. Our aim was to determine whether the decrease in frequencies of snow events belonging to each aforementioned pattern was due to climatic variability patterns or to an increase of temperature favoring increased melting level altitude. As already noted, despite the presence of patterns favorable for precipitation, there might be no snowfall at the stations.
We should first point out that measurements at weather stations can be easily affected by nonclimatic factors such as changes in instrumentation, exposure or measurement technique, observers, observation times, or the use of different methods to calculate monthly averages [
Subsequently, statistical techniques of the nonparametric Mann-Kendall test were applied at the 10 stations, to determine trends in number of snowfall days and their statistical significance. These techniques were applied using the MAKESENS template [
To assess this effect, we executed the method proposed by von Storch and Navarra [
The Mann-Kendall test is used in climate studies to detect trends in data series [
Finally, to estimate the magnitude of change per unit time of a linear trend, we used the nonparametric Sen’s slope estimator [
As noted above, applying PCA for geopotential height and temperature at 500 and 850 hPa, respectively, we extracted 5, 5, 8, and 7 components. These components were extracted until reaching 90% of explained variance for each variable. These extracted components were evaluated by means of the scree test [
Then, with a matrix of a total 25 PCA results (columns) and 241 study days (rows),
Figure
Cluster distribution of snow events by city between 1960 and 2011.
PCA applied to each variable. (a) Cumulative explained variance (%); black horizontal line indicates cumulative variance 90%. (b) Scree test. (c) North rule of thumb.
Synoptic environment of cluster 1. Geopotential height (HGT) and temperature (T) at 500 hPa (a, b) and 850 hPa (c, d).
As in Figure
As in Figure
As in Figure
Cluster 1 is defined by the arrival of maritime Arctic air over the Iberian Peninsula. An upper-air ridge associated with a surface high in the Atlantic, coupled with a trough in western Europe, produces a strong barometric gradient that causes horizontal advection of maritime Arctic air over the peninsula. With this configuration, a cold tongue at mid-levels settles over the northwest peninsula, whereas at low levels, the maritime flow generates significant cold and moist air advection over the northern peninsula (Figure
Cluster 2 is marked by the movement of storms across the North Atlantic, while the tropical ridge shifts to the southwest peninsula. Thus, an intense west-northwest barometric gradient that advects the Northwester Atlantic air is established over the peninsula. This air mass has a long maritime trajectory, acquiring high humidity and gradually warming at lower levels, owing to contact with the ocean. This is seen from 850 hPa temperatures (Figure
The northwest regions are in Castilla y León, and those at higher altitude are most affected by snowfall events since the melting level is relatively high. The cities of Ponferrada and León had conditions resulting in most snowfall events (Figure
The conditions described by cluster 3 correspond to arrival of a continental polar air mass over the Iberian Peninsula. A low-pressure system is situated over the western Mediterranean and an upper-air ridge over northwestern Europe. This forms a strong northwest barometric gradient that advects continental polar air. Very cold isotherms are evident under these conditions, at both 850 hPa and 500 hPa (Figure
Cluster 4 shows a cyclonic circulation at relatively low latitudes, affecting the Iberian Peninsula. By contrast, there is an anticyclonic circulation area in the North Atlantic, where there are typically passing disturbances (Figure
To determine the temporal evolution of snow events frequencies observed in each cluster, we plotted 10-year moving average frequency for the snow events recorded in each cluster. The linear regression trends were computed with the original unsmoothed data to avoid the serial correlation induced by the moving average smoothing. Figure
Ten-year moving average frequency of snow events. Cluster 1 (a), cluster 2 (b), cluster 3 (c), and cluster 4 (d).
The configurations describing the northeastern advection and Artic advection patterns had a very slight trend, without significant correlation at the 0.05 level. Thus, we cannot conclude that there were statistically significant changes in frequencies of these events. Therefore, in view of these results, we expect a decline in conditions favorable to snow precipitation in most of the cities analyzed. In particular, cities that correspond to the western flow pattern should show the greatest downward trend in number of snowfall days.
To extend the study of trends in snow days in each cluster, the nonparametric Mann-Kendall test was then applied to determine sign of observed trend and their statistical significance. The result of the Mann-Kendall test regarding the sign of the trend was −0.72 for cluster 1, −2.6 for cluster 2, 0.62 for cluster 3, and −1.33 for cluster 4. The trend was significant for cluster 2 at the 0.01 level. These results are consistent with the linear regression trends. A clear, significant, negative trend is observed for the western flow pattern. Meanwhile, trends in the northeastern and Arctic advection patterns are very weak and not significant. In addition, we analyzed seasonality of the snow days series by cluster. The autocorrelation function or correlogram was applied to the annual series of snow days by pattern, indicating no annual seasonality in any of the series.
Finally, some studies point to the connection between climate variability and teleconnection patterns, such as the North Atlantic Oscillation (NAO). This index uses the difference between the standardized subtropical Atlantic and subpolar Atlantic pressure values. Martín Vide and Fernández [
Table
Average NAO index on heavy snow events, by cluster and station.
Cluster | 1 | 2 | 3 | 4 |
---|---|---|---|---|
Ávila | −0.30 | −1.58 | 0.44 | −0.79 |
Burgos | 0.04 | −0.12 | 0.06 | −0.58 |
León | −0.22 | −0.08 | −0.50 | −1.23 |
Palencia | N.A. | −0.72 | N.A. | −0.65 |
Ponferrada | −0.82 | −0.34 | −0.97 | −1.39 |
Salamanca | −0.06 | −0.26 | −0.09 | −1.53 |
Segovia | −0.83 | 0.22 | −0.37 | −0.95 |
Soria | −0.17 | −0.18 | 0.30 | −0.80 |
Valladolid | N.A. | −0.28 | −0.09 | −1.70 |
Zamora | N.A. | −0.10 | −0.75 | N.A. |
Total | −0.15 | −0.05 | −0.05 | −0.88 |
Thus, a negative NAO favors the presence of a cyclonic circulation that produces precipitation across the study area. However, low temperatures are also necessary for snowfall. Therefore, we conclude that a strongly negative NAO greatly favors precipitation in the study area but not necessarily in the form of snow.
In this section, we examine the results obtained by analyzing trends in the number of snowfall days, maximum and minimum temperature, and precipitation.
The nonparametric Kruskal-Wallis test results indicate that all stations were homogeneous for all variables, except for the stations in the cities of Ávila (for snow days and maximum and minimum temperature) and Segovia (for maximum temperature). Serial correlation evaluation showed that the Burgos series had serial correlation for snowfall days and maximum temperature, the León station for minimum temperature, Ponferrada for snowfall days, and Salamanca for maximum temperature. In these cases, we performed prewhitening of the series prior to applying the Mann-Kendall test [
Table
Trends for maximum and minimum temperature, precipitation, and snow days for the stations studied, between 1960 and 2011.
Station |
|
|
Precipitation | Snow days | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Test |
Signific. |
|
Test |
Signific. |
|
Test |
Signific. |
|
Test |
Signific. |
| |
Burgos (Villafría) | 2.13 | * | 0.029 | 1.24 | 0.037 | −0.91 | −0.98 | −0.76 | −0.064 | |||
Valladolid | 2.47 | * | 0.029 | 2.42 | * | 0.056 | −1.46 | −1.546 | −1.58 | −0.05 | ||
Zamora | 3.56 | *** | 0.047 | 0.31 | 0.006 | −0.39 | −0.476 | −1.23 | −0.025 | |||
Soria | 2.33 | * | 0.024 | −0.24 | −0.006 | −0.53 | −0.614 | −1.23 | −0.069 | |||
León | 0.45 | 0.005 | 0.35 | 0.005 | −1.6 | −2.025 | −3.01 | ** | −0.142 | |||
Ponferrada | 3.14 | ** | 0.035 | −1.13 | −0.016 | −0.08 | −0.16 | −2.47 | * | −0.044 | ||
Salamanca | 3.21 | ** | 0.044 | −0.73 | −0.013 | −1.22 | −0.954 | −2.32 | * | −0.079 | ||
Palencia | 0.32 | 0 | 2.55 | * | 0.008 | 0.41 | 0 | NaN | ||||
Segovia | NaN | −0.81 | −0.018 | 0.54 | 0.477 | 2.032 | * | 0.08 | ||||
Ávila | NaN | NaN | 2.75 | ** | 2.247 | NaN |
Mann-Kendall test (
For precipitation, most of the stations had a downward trend except in Palencia, Segovia, and particularly Ávila, where the trend was significant at 0.01 level. Studies similar to ours by Gallego et al. [
Consequently, broadly considering the rising temperatures and declining precipitation, negative trends in the number of snow days are expected. Correspondingly, all stations except Segovia had negative trends of this number. The downward trend was stronger and more significant in the western cities of Castilla y León, especially León. This result is consistent with the first part of this study, which outlined synoptic patterns responsible for heavy snow in each city. Thus, heavy snowfalls in León were very dependent on the western flow synoptic pattern, which has significantly declined in frequency during the second half of the 20th century. In contrast, eastern cities of the region, like Burgos and Soria, had heavy snowfalls under a variety of synoptic patterns characterized by flows with a northerly component. These patterns have had slight trends in frequency, which coincide with more moderate declines in frequency of snowfall days. The city of Segovia stands out, where there was a positive trend in the number of snowfall days. This may be because, together with Ávila, it is situated on northern slopes where light snow typically occurs owing to lingering clouds. Thus, these snowfalls are recorded as snow days, but they are usually weak and therefore not represented in the heavy snowfall database.
These results suggest that a decrease in number of snow days at the analyzed stations can be attributable to rising temperatures and decreased frequency of patterns favorable for heavy snow. This is in line with the results of Pons et al. [
Knowledge of the types of synoptic scale patterns leading to a particular meteorological phenomenon can aid in understanding of the causes of climate variability and more precise explanation of its changes in trend. In this work, we classified synoptic scale patterns defined by temperature and geopotential fields at 500 and 850 hPa, respectively, for 241 days with recorded important snowfall in at least one of 10 main observation stations in Castilla y León, from 1948 to 2011. By applying PCA and
We obtained trends in number of snowfall days, maximum and minimum temperature, and precipitation for the stations from 1960 to 2011, using the Mann-Kendall test. The results show a significant decrease in number of snow days at stations in the northwest part of the region. We also found an increase in maximum temperatures and a decrease in precipitation although to a lesser extent. These results can be explained by conditions favoring snowfalls that dominate in an area. Thus, stations in the northwestern study area are very dependent on the western flow pattern for snowfall. This pattern is characterized by temperatures at lower levels that are not extremely cold. Consequently in recent years, there has been a pronounced decline in the number of snow events associated with this pattern. One possible cause is an increase of melting level, which has produced precipitation in the form of rain rather than snow at the stations.
However, cities in the eastern part of the region had the least negative trend for number of snow days. This area benefits from almost all the above patterns, especially those that generate advection with a northerly component. Thus, the decrease in frequency of snowfall events under these patterns was much smaller, because the melting level is low enough that a temperature increase fails to raise it above the station elevations.
Another possible cause of the decrease in number of snow days is an observed increase in the NAO index during winter months. Conditions conducive to snowfall at stations in the region are characterized by a neutral NAO, except for the cyclonic circulation pattern that has a clear negative NAO. Thus, at the Salamanca station, where this pattern occurs in more than 50% of snowfall events, this could be one cause of the significant decline in number of snow days.
Finally, only the Segovia station had a significant positive trend. The reason may be that it is the only station on the northern slope. Therefore, it commonly has lingering clouds from northerly advection, which produce snowfall days with only light precipitation.
As we have shown, changes in trends of meteorological phenomena respond to variations in large-scale atmospheric circulation patterns. In the special case of the snowfalls, they are directly affected by increased temperatures, and it is difficult to establish how much the reduced snowfall may be attributed to rising temperatures and how much to changes in weather patterns. In-depth study of the behavior of these patterns provides a basis for future analysis of trend changes in meteorological variables.
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors would like to thank the Junta of Castilla y León (LE220A11-2 and LE176A11-2) and the CGL2010-15930 for their support. Additionally, they wish to thank the State Agency of Meteorology (AEMET) for the use of data from their weather station networks. Sergio Fernández-González acknowledges the support from the FPU program (AP 2010-2093).